People & Strategy Summer 2017 Vol. 40 Issue 3 - 39
attempts. Mergers, acquisitions, organizational changes, staffing changes, computer upgrades, and data retention policies
can all make it more difficult, if not impossible (or, in the case
of retention policies, inappropriate) to reconstruct historical
Some designs will require that new data be generated.
Here, standard quantitative and qualitative techniques such as
surveys and focus groups are essential tools. Emerging methods that gather passively generated data are also increasingly
used. This may include online behavior such as movement
through a job application process, intranet query behavior, or
interactions with customers or colleagues via email, calendar
or chat. It may also include video and audio encoding of
interactions such as interviews or manager coaching sessions
with employees, or real-life tracking via a wearable device that
can track location, proximity to others, how much the wearer
is talking during meetings and conversations, how much time
individuals spend sitting, and so forth.
These new data streams open up interesting new possibilities to breathe new insight into old questions. For example,
measurement of social behavior can bring new light to team
and task design that contribute to team effectiveness. Here,
the business decision that would be influenced might be
about process and team design.
To return to our longitudinal study predicting the best
hires, useful data might include resume at time of hire,
interviewer feedback, first manager, indicators of manager
quality, onboarding buddy or trainer, indicators of the quality
of that onboarding buddy or trainer, starting job salary, and
position in the salary range. Outcome measures might include
things like several years of performance data, salary velocity,
promotions, assessment results, awards and recognitions, and
whether the person is still with the organization, or, if they
left, their total tenure.
Artificial intelligence can consider thousands of variables,
and HRIS and applicant tracking systems (ATS) certainly have
oceans of data available, including such things file extension
of the resume, browser used to access the online job portal,
direct deposit information, number of dependents, and zip
code. However, talent analytics is somewhat distinct from
other fields in that labor law places restrictions on the types of
data that can be used for employment decisions. For example,
data used in employment decisions must be job-relevant. Even
if variables such as the file extension of the resume are useful
predictors of success, they are not appropriate to use.
Step 4: Effectively and Appropriately Analyzing the Data
This step is rightly the center of the process. None of the
other steps matter if the actual analysis is not done correctly.
Analyses broadly fall into three main categories: descriptive,
predictive, and prescriptive. As the names suggest, descriptive
analyses address the current state, predictive analyses reveal
likely future outcomes, and prescriptive analyses recommend
analyses. In general, effectively and appropriately analyzing
the data is a strength area for talent analytics teams and professionals; thus, this section will provide a lighter overview.
The advent of data visualization has breathed new life into
descriptive analyses. Data visualization can enable a quick
and sophisticated understanding of a complex set of relationships among variables and over time. Data visualization is also
frequently dynamic, allowing researchers and end users the
ability to explore the data, such as the ability to "drill" into a
specific cell or relationship.
A capable analyst will be proficient in multiple methods
and able to use multiple platforms and analytic packages. Predictive analyses have long been considered the gold standard
for talent analytics in areas such as selection. Predictive analytics, as the name suggests, outline what is likely to happen
in the future. Prescriptive analytics are relatively new to talent
analytics practitioners. Prescriptive analytics are different from
predictive in that prescriptive analytics tell you what to change
to achieve the outcomes that you desire.
A key risk for talent analytics
functions is in delivering insights that
are interesting, but not actionable.
Often, the most time-consuming part of an analytic process
is preparing the data. Often, data have to be reformatted,
joined together from multiple sources, or have discrepancies
between databases corrected. For example, one database
might have a job title of "sr. financial analyst" while another
in the same organization spells out "senior financial analyst."
Humans can easily identify that those two things are the same.
Computers have to be told.
Step 5: Developing Insight Based on the Analyses
Too often, researchers simply share the full results of their
analyses. However, in Step 1, we identified a business problem. Volumes of analyses can do more harm than good in
that they may obscure interesting findings. The purpose of
analysis is not to produce reams of results, but rather to synthesize results into an insight that can drive action.
Several factors can get in the way of developing this insight.
The first is framing. Junior researchers in particular may think
that their job is to run analyses, and it is executives' job to
figure out what to do with it. However, a significant portion of
the value of a talent analytics function is this ability to synthesize a complex set of findings into a compelling insight.
The second challenge is insufficient understanding of the
key business problem being addressed. This is part of why
Step 1 is so critical. If the treatment of the problem is limited
to the key variables and does not extend to the business problem or the processes being examined, then the ability to identify the key findings is limited. Step 1 is also critical because if
my questions did not cover enough of the domain, my ability
to develop a credible insight will be limited.
The recommendation in Step 4 to run a variety of analyses can be important for insight generation as well. Multiple
perspectives on a problem can offer opportunities for nuance.
However, perhaps paradoxically, helping facilitate insight
and action in a group of leaders or managers is often easier if
results are simplified, or only a subset of analyses are shared.
VOLUME 40 | ISSUE 3 | SUMMER 2017